Connect to a model provider

Last published : Nov 09, 2026
This guide shows you how to sign the assistant in to a model provider, and how to reconnect, test, or sign out.
The assistant needs a signed-in model provider to answer. Until one is connected, Knowledge Worker Agent opens Set up your assistant when it starts, and the chat shows "The assistant isn't connected." with a Connect button.
Related information
Use Knowledge Worker Agent

Open your accounts

  1. At the bottom of the sidebar, select Settings (the gear icon).
  2. Select Accounts.
The Accounts & connections dialog lists each model provider your workspace offers, with its status.
Which providers you see depends on your workspace. Your administrator may assign them centrally. If the dialog shows No model provider configured, contact your administrator.

Use the setup guide

Until the assistant has a model that works, Set up your assistant opens each time Knowledge Worker Agent starts. It shows two steps:
  • Connect a model: Set up a model opens Accounts & connections. If there's nothing you can do yet, the step says so instead: "No model provider is assigned to this workspace. Contact your administrator.", or "Your administrator is still setting up" and the provider's name.
  • Connect your apps: the connectors you haven't set up yet, each with what it's for. Open connectors opens Connectors.
When you're done in Accounts & connections or Connectors, you're back in the guide, which marks what you set up.
Close closes the guide until Knowledge Worker Agent starts again. To stop it from opening, select Don't show this again before you close it. The message "The assistant isn't connected." stays in the chat until a model works.
When a model already works and your administrator assigns a model provider you haven't set up, New in your workspace opens instead, the next time Knowledge Worker Agent starts. For more information, see Manage connectors.

Sign in to GitHub Copilot

  1. Next to GitHub Copilot, select Connect.
  2. Under Copy this one-time code, select Copy.
  3. Under Open GitHub and paste the code, select the button. GitHub opens in a new tab.
  4. Paste the code in GitHub and approve the request.
  5. When "✅ Connected! The assistant is ready to use." appears, select Done.
If sign-in fails, select Retry. To see what went wrong, expand Show CLI output.

Sign in to Google Vertex AI

This connects the assistant to Anthropic Claude on Google Cloud Vertex AI, using your Google account.
Note: Your administrator must first set the Google Cloud project for Vertex AI in the workspace. Without it, Vertex AI can't be used. For the setting, see Environment variables.
  1. Next to Google Vertex, select Connect.
  2. Select 🔗 Open Google sign-in, and sign in to Google in the new tab.
  3. Copy the verification code Google shows.
  4. Back in Knowledge Worker Agent, paste the code into Verification code, and select Submit.
  5. When "✅ Connected! Vertex AI is ready to use." appears, select Done.

Add an API key

Some providers that your administrator assigns ask each person for their own key.
  1. Next to the provider, select Add key.
  2. Paste your key into Paste your API key.
  3. Select Save.
Your key is stored in your workspace only, and is used only to reach that provider. To change it, select Replace key. To delete it, select Remove.

Providers your administrator manages

A provider marked managed is set up by your administrator. There's nothing to sign in to. If it isn't ready yet, the chat shows "Your administrator hasn't finished setting up" followed by the provider's name.

Test a connection

Next to a connected provider, select Test. The status updates with the result.

Reconnect or sign out

  • To sign in again, for example with a different account, select Re-authenticate.
  • To sign out, select Sign out, then Confirm.
Signing out of a provider that doesn't work for you stops the assistant from offering its models.